Tokenomics: Why making AI pay is tricky
📰 source: bbc_business · 💼 business
figuring out how to charge for ai is a mess. companies like microsoft, google, and anthropic have spent hundreds of billions building llms, but the economics of tokens — the chunks that make up prompts and responses — are unpredictable. the same prompt can give different outputs, and agentic systems burn through tokens in ways nobody can forecast.
goldman sachs says token consumption will rise 24 times between 2026 and 2030 to 120 quadrillion a month. meanwhile, uber tore through its annual coding token budget in months, and microsoft reportedly reined in engineers' use of third-party coding tools. businesses don't know how many tokens they're using until the bill arrives.
experts say pricing is guesswork. saviynt's simon gooch says locking customers into multi-year cost models makes no sense. some firms use flat-fee personal accounts, but smartr ai's oliver king-smith says that will end when vendors face shareholder pressure. options like per-result pricing or bundles are on the table, but sumo logic's bill peterson notes vendors change prices every few months, so nobody can budget.
why it matters: unpredictable token costs make ai pricing a moving target, forcing vendors and customers to rethink budgets and revenue models.
source: bbc_business
sentiment: -0.10 · impact: 0.30